JPH0981615A - Circuit designing device and method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To design a large scale logic circuit. SOLUTION: Based on genetic algorithm, the circuit constitution of a programmable logic circuit (PLD) is updated and the logic circuit for performing object output is designed. Corresponding to the genetic algorithm, the set of grammatical rules (for instance, (a) to (d)) for leading out the circuit constitution of the PLD is turned to a chromosome, the chromosome (the set of the grammatical rules) is updated and the chromosome for supplying optimum circuit constitution is led out. At the time, since the length of the chromosome is proportional to the number of the grammatical rules, without depending on the scale of the circuit of the PLD, the circuit constitution is designed in appropriate calculation time even when the circuit of the PLD is a large scale.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a circuit designing apparatus and method, and more particularly, based on a genetic algorithm,
The present invention relates to a circuit design device and method for changing a circuit configuration.

[0002]

2. Description of the Related Art Conventionally, when designing a complicated logic circuit for controlling a robot or the like, a designer has empirically combined many basic logic elements to design the logic circuit. However, recently, genetic algorithms (G
A: A method of designing a logic circuit based on the Genetic Algorithm) without requiring empirical knowledge is described in, for example, “History of Genetic Algorithm (Hiroaki Kitano, Sangyo Tosho)” in “Genetics”. Basic experiment of hardware evolution by dynamic learning ”is introduced.

In this method, the FPGA (Field Pr
Programmable Logic Device (PLD) represented by ogrammable gate array)
The circuit configuration of (1) is repeatedly changed based on GA to create a logic circuit that outputs a desired signal.

The PLD has a plurality of logic cells capable of dynamically selecting a basic type of logical calculation such as AND logical operation and OR logical operation, and the type of logical calculation executed by each logical cell. , The pattern of circuit coupling between these logic cells can be changed.

In GA, what characterizes the object to be optimized is expressed by a gene, and by connecting this gene, a chromosome is generated. Then, by repeatedly updating a plurality of chromosomes, the target is brought closer to the optimum state.

Conventionally, the function of logic cells in PLD and the pattern of circuit connection between logic cells are represented by chromosomes,
This chromosome (PLD circuit configuration) is updated on the basis of GA to create a logic circuit for performing a desired output.

[0007]

However, when designing a logic circuit that performs complicated control, a large number of logic operation elements are required, and the coupling pattern between these logic operation elements also becomes complicated. In the conventional technique, the function of the logic cell and the pattern of circuit connection in the PLD are expressed by the chromosome, and thus the length of the chromosome becomes longer according to the number of the logic cells.
Therefore, when designing a large-scale logic circuit, it is necessary to perform an operation based on GA for a very long chromosome, which causes a problem that a lot of calculation time is required.

The present invention has been made in view of such a situation, and a function of a logical cell and a connection pattern are generated by a set of grammar rules consisting of a predetermined number of grammar rules, and these grammar rules are set in a chromosome. By expressing with, the length of the chromosome does not depend on the number of logic cells, and a large-scale logic circuit can be designed.

[0009]

According to a first aspect of the present invention, there is provided a circuit designing apparatus, wherein an arithmetic means capable of performing an arithmetic operation and dynamically changing the arithmetic function and a grammar rule for deriving the arithmetic function are chromosomes. Control means for changing the arithmetic function so that the output of the arithmetic means approaches the target output based on the genetic algorithm.

According to a second aspect of the present invention, there is provided a circuit design method based on a genetic algorithm having a chromosome as a grammatical rule for deriving the arithmetic function of an arithmetic element capable of dynamically changing the arithmetic function for performing an arithmetic operation. The arithmetic function is changed so that the result of the arithmetic approaches a target value.

In the circuit design apparatus according to the first aspect, the arithmetic means capable of dynamically changing the arithmetic function performs an arithmetic operation, and the control means causes the genetic rule whose chromosome is a grammatical rule for deriving the arithmetic function. Based on the genetic algorithm
The calculation function is changed so that the output of the calculation means approaches the target output.

In the circuit designing method according to the second aspect, based on a genetic algorithm having a chromosome as a grammatical rule for deriving the arithmetic function of an arithmetic element which performs an arithmetic operation and can dynamically change the arithmetic function. Then, the calculation function is changed so that the result of the calculation approaches the target value.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a configuration example of an embodiment of a circuit designing apparatus of the present invention. This configuration example is a PLD
1 (computing means). The PLD 1 is connected to the arithmetic unit 2 (control means), and according to the control signal of the arithmetic unit 2,
By changing the circuit configuration, the robot 11
The control signal is output to.

The robot 11 is adapted to operate according to the control signal supplied from the PLD 1. The designer 21 observes the operation of the robot 11, evaluates the operation according to a predetermined evaluation method (evaluation function), operates the input device 3 and inputs the operation device 2 corresponding to the evaluation. .

The arithmetic unit 2 holds a predetermined number of chromosomes, derives the circuit configuration of the PLD1 from each chromosome, and outputs it as a control signal to the PLD1.
Further, after the operation of the robot 11 and the evaluation of the designer 21 are completed for all the chromosomes, the arithmetic unit 2 determines that the descendants of the chromosomes corresponding to the circuit configuration realizing the highly evaluated operation are the next generation based on the GA. It is designed to update the chromosomes with a high probability of remaining.

FIG. 2 shows a configuration example of the PLD 1.
This configuration example includes a plurality of logic cells 41-1 to 41-N. These logic cells 41-1 to 41-N are
It has a plurality of basic logic calculation functions such as ND logic operation and OR logic operation, and dynamically changes the type of logic calculation to be executed according to a control signal supplied from the outside.

The logic cells 41-1 to 41-N are designed so that any logic cells can be connected to each other. Whether or not to connect the logic cells to each other is determined by a control signal supplied from the outside. Is determined by. When coupling between logic cells, the corresponding coupling switch among the coupling switches 42-1 to 42-M is turned on. For example, when the logic cell 41-N and the logic cell 41-1 are coupled, the coupling switch 42-1 is turned on.

FIG. 3 shows a configuration example of the arithmetic unit 2. This configuration example includes a CPU 61, and the CPU 61
According to the program stored in the ROM 62,
According to various processes, for example, a program of GA, the designer 21 operates the input device 3 and calculates a new circuit configuration of the PLD 1 according to the evaluation of the operation of the robot input. The RAM 63 is configured to appropriately store data, programs, etc. necessary for the CPU 61 to perform various processes.

The evaluation of the motion of the robot 11 input by the designer 21 is input from the input device 3 via the interface 64. Further, the output of the control signal to the PLD 1 is also performed via the interface 64.

Next, referring to the flowchart of FIG.
The operation of the above embodiment will be described.

First, in step S1, the arithmetic unit 2 creates a predetermined number of grammar rules. As shown in FIG. 5, each grammar rule has a 2 × 2 matrix in which one alphabet is on the left side and each element is an alphabet on the right side. The computing device 2 creates a gene from this grammar rule and connects a predetermined number of genes to form a chromosome. The arithmetic unit 2 generates a predetermined number of initial chromosomes.

For example, when the rule 1 in FIG. 5A is a gene, the rule 1 is expressed as AOCHL. Therefore, for example, the rule 1 of FIG. 5 (a), the rule 2 of FIG. 5 (b), and the rule 3 of FIG. 5 (c) are expressed by genes,
When the chromosomes starting with these genes are described, it becomes AOCHLFKHLKHTQJT ... As shown in FIG. In this way, a predetermined number of chromosomes of a predetermined length are created.

Next, in step S2, the logic cell 41-
Generate an alphabetic matrix with a size corresponding to the numbers 1 to 41-N. For example, as shown in FIG. 7A, first, one alphabet (O) is set,
Grammar rules that have this alphabet (O) on the left side,
For example, the rule 4 shown in FIG. 5D is applied to convert the alphabet (O) into the 2 × 2 matrix shown in FIG. 7B. Then, for each element Q, O, T, and E of this 2 × 2 matrix, by applying a grammar rule (not shown) having these alphabets on the left side, this 2 × 2 matrix becomes It is converted into the 4 × 4 matrix shown in FIG.

Similarly, when the grammatical rule is applied to each element of this 4 × 4 matrix, the 8 × 8 matrix shown in FIG. 7D is obtained, and when the grammatical rule is applied to the elements of this matrix, FIG. It becomes the 16 × 16 matrix of (e). Thus, the number of rows (= number of columns) in the alphabet matrix
Applies the grammar rule to each element of the matrix until the number of logic cells 41-1 to 41-N in PLD1 is greater than or equal to. Therefore, it is possible to create a large-scale alphabet matrix with only a predetermined number of grammar rules.

For example, when the number of logic cells 41-1 to 41-N is 16 (N = 16), the size of the matrix shown in FIG. 7 (e) is sufficient, and the matrix of this alphabet is converted. Then, a logic circuit coupling matrix representing the coupling state of the logic cells and the function executed by each logic cell is created.

The diagonal elements of the logic circuit connection matrix represent the functions of the logic cells 41-1 to 41-N, and the right upper half elements of the logic circuit connection matrix are the connection states of the logic cells 41-1 to 41-N. Express For example, the i-th logical circuit coupling matrix
If the components of the row and j-th column are represented by Aij, as shown in FIG. 8, when the logic cells 41-1 to 41-N have a total of 6 functions, the diagonal component Aii is an integer of 0 to 5. And the function of the i-th logic cell 41-i (AND logic operation, O
R logical operation).

The off-diagonal component Aij is 0 or 1
Then, the connection state of the i-th logic cell 41-i and the j-th logic cell 41-j is expressed. When Aij = 1, Aij indicates that the logic cell 41-i and the logic cell 41-j are combined, and when Aij = 0, Aij is Aij.
Indicates that these logic cells 41-i and 41-j are not connected. Therefore, the coupling state can be expressed only by the upper right half or the lower left half of this matrix. In this configuration example, the upper right half component of the matrix is used.

Next, to create this logical circuit coupling matrix from the alphabet matrix as shown in FIG. 7E, for example, the diagonal elements of the alphabet matrix are 0, 1 in alphabetical order from A. , 2, 3, 4,
5, 0, 1, 2, 3, ..., And 0 to 5, and for non-diagonal components, A, B, and C are converted to 0, and other alphabets are converted to 1. As a result, the alphabet matrix shown in FIG. 7E can be converted into the logic circuit coupling matrix shown in FIG.

In this way, the arithmetic unit 2 generates a logic circuit connection matrix from the set of grammar rules, and outputs this matrix to the PLD 1 as a control signal. The PLD 1 has a circuit configuration (logic cells 41-1 to 4-4) according to the control signal.
1-N functions and binding states).

Next, in step S3, the PLD 1
According to the logic circuit set in step S2, the robot 1
1 is operated. The designer 21 observing the operation of the robot 11 evaluates the operation according to a predetermined evaluation method in step S4, operates the input device 3, and inputs the evaluation to the arithmetic unit 2.

In step S5, the arithmetic unit 2 determines from the evaluation input by the designer 21 whether or not the robot 11 has performed an operation satisfying the designer 21, and the designer 21
If the designer 21 is not satisfied, the process proceeds to step S6.

In step S6, it is determined whether or not the robot 11 has been operated for all chromosomes, and the processes of steps S2 to S5 are repeated until the robot 11 is operated for all chromosomes. After the processing of (1) is completed, the process proceeds to step S7.

Then, in step S7, the arithmetic unit 2 performs three processes of selection process, crossover process, and mutation process based on GA based on the evaluation of the designer 21 with respect to the operation of the robot 11 for each chromosome. Generate the next generation of chromosomes.

In the selection process, a chromosome is selected from a group of chromosomes with a probability proportional to the evaluation of the designer 21, and a pair of chromosomes is created. Therefore, there is a high probability that many pairs will be created for a highly evaluated chromosome (which realizes the operation of the robot 11), and a high probability that many descendants will be left in the next generation.

In the crossover process, for each pair selected in the selection process, a place where two chromosomes are crossed is determined by a random number, and the values of all the digits after that digit are determined between the two chromosomes. Replace with. For example, at the 4th digit from the left, the chromosome AO
CHLFKHLK ... and chromosome BTTDTCDMTP
When crossing ..., chromosome BTTDLFKHLK
, And two chromosomes of the chromosome AOCHTCDMTP ...

The mutation processing changes the alphabet of digits determined by a random number in a chromosome, and is performed with a certain low probability when creating a next-generation chromosome. When performing mutation processing, the digit to be changed is determined using a random number and the alphabet of that digit is changed. For example, on the chromosome JFAIEMJECP, 3 from the left
When a mutation occurs in the digit (alphabet A), the treated chromosome is, for example, JFBIEEMEJCP.

After generating the next generation chromosome, step S
Returning to step 2, the processes of steps S2 to S7 are repeated until the designer 21 is satisfied with the operation of the robot in step S5.

As described above, a set of grammatical rules for deriving the circuit configuration of the PLD 1 is set as a chromosome, and the grammatical rule (chromosome) is updated based on GA to design a logic circuit that outputs a target signal. .

FIG. 9 shows the relationship between the fitness and the generation of the chromosome when solving an example of the MXOR (multiple exclusive OR) problem in one embodiment of the circuit design circuit and method of the present invention. For the fitness, the output of the PLD represented by each chromosome is
It is an index showing how close the output is to the target. This MXOR problem aims to realize a relationship between inputs and outputs (16 inputs, 8 outputs) of a logic circuit composed of 8 XOR logic elements by using a PLD having 64 logic cells. .

A predetermined input is applied to the PLD, and the GA is used to reduce the difference between the output of the PLD for this input and the output (solution of the problem) when eight XORs are used. The function will be updated.

In this example, when the conventional technique is used, the fitness is 750 (maximum value) even in the 95th generation, whereas one embodiment of the circuit designing apparatus and method of the present invention is used. In this case, in the 10th generation, a chromosome expressing a circuit configuration having a fitness of 780 appears, and the circuit can be designed quickly.

FIG. 10 shows an embodiment of the circuit designing apparatus and method of the present invention, which is a 6-multiplexer.
It shows the relationship between fitness and the generation of chromosomes when solving an example of a problem.

The 6 multiplexer has four input channels, two multiplex signal channels and one output channel, and outputs the value of one of the four input channels according to the value of the multiplex signal channel. Output to. The 6-multiplexer problem is to realize the relationship between the input and the output of the 6-multiplexer.

In this example, the PLD is provided with four input channels and inputs corresponding to two multiplex signal channels, the output of the PLD for this input and the output of the actual 6 multiplexers (solution to the problem). The GA is used to update the function of the PLD so that the difference becomes smaller.

When the conventional technique is used, the fitness is 0.66 (maximum value) even in the 45th generation, whereas when the embodiment of the circuit designing apparatus and method of the present invention is used, 5
In the generation, a chromosome expressing a circuit configuration having a fitness of 0.98 appears, and the circuit can be designed quickly.

Although the control circuit of the robot 11 is designed in the above embodiments, the present invention can be applied to the design of a general logic circuit.

In the above embodiment, the entire area of the logic circuit is designed by utilizing the GA, but by converting the already designed logic circuit into grammar rules and incorporating the grammar rules into the GA chromosome, The logic circuit can be used as part of a newly designed logic circuit. In this case, the crossover process and the mutation process in the GA are not performed on the incorporated grammar rule, and the designed logic circuit is incorporated into the newly designed logic circuit.

[0048]

As described above, according to the circuit designing apparatus of the first aspect and the circuit designing method of the second aspect,
Based on a genetic algorithm in which the grammatical rule that derives the arithmetic function is a chromosome, the arithmetic function is changed so that the result of the operation approaches the target value. Since it is proportional to the number and does not depend on the scale of the logic circuit, a large-scale logic circuit can be designed quickly and easily.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of an embodiment of a circuit design device of the present invention.

FIG. 2 is a block diagram showing a configuration example of a PLD 1 in the embodiment of FIG.

FIG. 3 is a block diagram showing a configuration example of an arithmetic unit 2 in the embodiment of FIG.

FIG. 4 is a flowchart illustrating the operation of the embodiment of FIG.

5 is a diagram showing an example of grammar rules used in the embodiment of FIG.

FIG. 6 is a diagram showing an example of a chromosome used in the embodiment of FIG.

FIG. 7 is a diagram showing an example of expanding a matrix according to the grammatical rule as shown in FIG.

FIG. 8 is a diagram showing an example of a logic circuit coupling matrix derived in the embodiment of FIG.

FIG. 9 is a diagram showing the relationship between fitness and chromosome generation when an example of the MXOR problem is solved by applying an example of the present invention.

FIG. 10 is a diagram showing the relationship between fitness and chromosome generation when an example of the 6-multiplexer problem is solved by applying an embodiment of the present invention.

[Explanation of symbols]

 1 PLD 2 arithmetic unit 3 input device 11 robot 21 designer 41-1 to 41-N logic cell 42-1 to 42-M coupling switch 61 CPU 62 ROM 63 RAM 64 interface

Claims (2)

[Claims] 1. An operation means capable of performing an operation and dynamically changing the operation function, and an output of the operation means based on a genetic algorithm having a chromosome as a grammatical rule for deriving the operation function. And a control means for changing the arithmetic function so as to approach the output of the circuit designing device. 2. The result of the operation is based on a genetic algorithm having a chromosome as a grammatical rule for deriving the operation function of an operation element that performs the operation and can dynamically change the operation function. A circuit designing method, characterized in that the arithmetic function is changed so as to approach the above.